larroy commented on issue #13409: [MXNET-1234] Fix shape inference problems in 
Activation backward
URL: https://github.com/apache/incubator-mxnet/pull/13409#issuecomment-443335902
 
 
   Do you guys have a better suggestion? I don't think with the current 
architecture there's a better solution, as there's no context information in 
functions like:
   
   mxnet::op::ActivationGrad::operator() activation.cc:148   to know if we are 
in CPU or GPU context.  As MKLDNN, CUDNN and CPU have all different set of 
inputs for Activation Backward, I don't see another possibility other than this 
fix. It's also a bit of a corner case having a CPU activation backward running 
when CUDNN is available, I don't see the concern when most users won't have 
such a case. When you have a GPU available why would you run it on the CPU?

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to